[llvm] [AMDGPU] Extend llvm.amdgcn.update.dpp intrinsic to support f64 (PR #91190)

[Jay Foad via llvm-commits]

jayfoad wrote:

LGTM, though I wish there was a better way to implement it then duplicating the selection pattern for every 64-bit type.

Is there any reason the sdag and globalisel tests can't be combined into a single file with two RUN lines?

https://github.com/llvm/llvm-project/pull/91190